Longitudinal Data Based Research on Web User Interests Drift Modeling

Abstract

User interests drift modeling is the fundamental and key technology of Personal Recommendation System (PRS), and is arising more and more attention by researcher from home and abroad. At present, web user interests drift model is in its original phase for the reason of many aspects, and there also exist some problems need to be studied continually and deeply. Considering of web user interests might change with time flying, huge of surfing data and many influence factors need to be concerned, because user interests changes constantly with time flying. A novel web user interest drift pattern modeling method is proposed. First, this paper takes user interests as time series via using hidden Markvo model, which may properly map the sequential characteristics of user interests. Second, improved GSP searching algorithm is used to find the frequent pattern from user interest sequence. Finally, forgetting mechanism is used to solve interests drift problem, and time window is used to store current user interest pattern, and the old interests will elapse with the new interests coming constantly. Plenty of online experiments were done to verify the reasonability of the user interests drift model. Via experiments, conclusion can be drawn that this method can work perfectly and efficiently, which may firmly grasp user interests drift rule in time. Therefore, personal recommendation system using these interests drift model can recommend the latest information for web user, only according to user's potential interests changing.